Prediction Of Blood Glucose Based On Six Statistical Learning Methods And Adaboost Perspective

Diabetes is a chronic disease that can be controlled but not cured.If we can reasonably use some statistical methods to predict the blood sugar value,it will not only help patients with high blood sugar value to take timely measures to control blood sugar value,but also effectively reduce the number of people suffering from diabetes and hyperglycemia,which has an important contribution to the improvement of the overall physical quality of our people.In this paper,in the process of predicting blood glucose,a total of 6 different statistical learning methods are used to predict the blood glucose value,that is,principal component analysis(PCA),gradient boost decision tree(GBDT),support vector regression(SVR),nuclear ridge regression(KRR),Adaboost integration,and VotingRegressor,and formed 6 integration models.The data on blood glucose values comes from the Tianchi Precision Medicine Contest-artificial intelligence assisted genetic risk prediction for diabetes.First,preprocess the data,import the processed data into Python,and then randomly divide a set of blood glucose data containing5642 sample values into two groups at a 7: 3 ratio,called the training set and the test set,and finally use The data in the training set uses 6 statistical learning methods to establish a regression model,the data in the test set is used to predict the blood glucose value,and the model is tested.At the end of the article,the six integrated models are compared and analyzed from the aspects of model accuracy and model efficiency.It is found that the Ada-VotingRegressor model has the highest accuracy,the mean square error of the training test set and the training set is relatively minimal,the difference between the mean square error of the test set and the training set is the smallest,the model is simple,and the fit is high;but when the model efficiency is considered,the model efficiency of the PCA-GBDT model is much higher than the other five models.